2,232 karma · joined April 25, 2016
The closest thing that comes to mind is to use a unique AES key for every chunk of 5 minutes or so, and then encrypt those keys with an asymmetric cipher with the private key secured off-site somewhere, which could selectively decrypt some of the AES keys. But that could be an inconveniently long list of keys for, say, 10 video feeds over several days.
The pandemic and trade wars put a wrench in the gears of that plan, but I still think that would be way cooler than lugging around a 128GB laptop and using it for everything, a.k.a. single point of failure.
Sure, a teacher with zero knowledge of chemistry can follow a syllabus, read the textbook out loud, make sure everyone's multiple choice answers match the answer key, and try to teach chemistry to some first approximation of "teach". A primitive computer program can do that too.
What happens when a student asks a question that deviates slightly from the syllabus because they didn't quite grasp it the way it was explained? The teacher can't answer the question, but a "dumb GPT model" trained on the entirety of the Internet, including a ton of chemistry, including the syllabus, probably can.
But yes, if you pre-filter the training data to include only the words of the syllabus, the language model will be just as poor of a teacher as the human who did the same thing. Reminds me of my first "programming teacher", a math teacher who picked up GW-BASIC over the summer to teach the new programming class. He knew nothing.
We never even reached the contentious part of this discussion. This part should be obvious. You can't just filter out "bomb" because the student could explain in broad terms what they mean by the word and then ask for a more detailed description. You can't filter out "explosion" because the teacher might need to warn the students about the dangers of their Bunsen burner. You can't filter out all the possible chemical reactions that could lead to an explosion, because they can be inferred from the subject matter that it's supposed to be teaching.
The same goes for things like negative stereotypes, foul language, talking about "illegal or unethical" behaviors (especially as laws and ethical norms can change after training, or differ between usage contexts). Pre-filtering is just a nonstarter for any model that's intended to behave as an intelligent being with worldly knowledge. And if we drop those requirements, then we're not even talking about the same technology anymore.
If so, I wonder about the implications of black-box models helping to study and develop solutions for interpretability. It gives me "reflections on trusting trust" vibes.
... oxidation, exothermic reactions?
A teacher who never actually studied the subject and only has knowledge of the pre-filtered material from the textbook would be little more than a talking textbook. I mean, textbooks are useful, but we already have those.
> Without a recipe or off-the-shelf explosives?
If they've ever studied the relevant subjects, yes.
> And even if true, a teacher who hasn't put themselves to the task of building a pipe bomb can be a perfectly good chemistry teacher.
But the suggestion to pre-filter is more like a teacher who was never allowed to learn enough to succeed at the task even if they tried. Teachers who could succeed but refuse to try are the norm. (Or maybe they have tried on their own time, or done the thought experiment, but they know that handing out step-by-step instructions would be taboo in most circumstances.)
> These models can't figure anything out, they just regurgitate and associate on command.
I think you may be overemphasizing "regurgitate" and underemphasizing "associate". ChatGPT can emit instructions for just about anything that were clearly not just regurgitated from some other source, and surprisingly often they're even accurate - or accurate enough to use as a starting point.
Let's just agree that pre-filtering the training text is not a practical way to prevent these models from producing emissions that are considered dangerous or taboo, at least not without losing the traits that make them so interesting in the first place.
But the self-censoring hidden prompts that they're using can also lead to absurd results, so there's a lot of room for improvement. I was trying to generate a fictional story where at one point the main character tries walk out of a store before paying for an item, and it refused to write something "illegal or unethical". It was okay with having the character leave some money on the counter before walking out.
I think if you go into class and ask the teacher how to build a bomb, they will refuse to answer, but not because they were never allowed to learn enough about the world to figure it out for themselves.
Just about anyone with a basic education can figure it out if they're put to the task, but that doesn't mean the parents and school administrators want it taught in the classroom. The teacher also had to learn the concept of "a device that explodes with great force and is typically used to harm other people" somewhere in order to recognize that they're not supposed to present those kinds of instructions in general.
And this is just an example. The same goes for thousands of other topics in hundreds of other fields that would be inappropriate for teachers to teach. If you shelter them all from the world to the point that they're completely ignorant of the forbidden topics, they would be ineffective as teachers. A school full of those teachers would be ineffective as a school. A society filled with such schools would be stunted as a society.
If you don't want it explaining how to build pipe bombs, either the filtering has to come later or you have to accept that the model will be useless for tons of innocent things that users would obviously want it to do.
It's one thing to narrow down a city-sized population to a list of 50 people known to walk with a similar gait, and then follow up with other forms of investigation before drawing conclusions. It's another for a cop to look at the top match and go arrest that guy, as they've done more than once with facial recognition.
Normal private property like someone's quadcopter, I would assume the owner still has rights to it (and possibly liability depending on how/where it fell). But foreign government property that's not supposed to be there in the first place?
Are you sure about that? I'm pretty sure I recall being able t&%@^0158
NNNNNNNN@*^&
NO CARRIER
Never said it wasn't.
> Plenty of countries managed to effectively flatten the curve regardless of the availability of N95 masks
"Flatten the curve" is a post-failure strategy to minimize secondary losses due to overwhelming the hospitals until we have a vaccine with enough efficacy to reach herd immunity and eradicate the virus. That latter part never happened.
> Statistical analysis showed that the odds of infection were about half for people who reported wearing a mask in public compared with people who didn’t.
Sure, and that's not even limiting the scope to N95-level masks.
> For people who wore masks “all of the time” (instead of “some of the time” or “most of the time”) the estimated effect was even more significant.
Naturally.
> Not unexpectedly, N95/KN95s were found to reduce the odds of infection compared with people who didn’t wear any mask. To me, the surprising thing is how effective they were, reducing the relative odds by 83%. Cloth masks and surgical masks were found to be less effective.
Sounds about right.
> All masks provide some non-trivial amount of protection. Certainly better than not doing anything, especially if you do so consistently.
Never said they didn't.
> Except I’ve been at parties where people with COVID showed up and no one else got sick.
Useless anecdote.
> And this is anecdotal and not quantitatively scientific.
Correct.
> I don’t know what your threshold is for “as it needs to be”.
Enough to reduce the effective "R" rate of transmission to below 1 and to keep outbreaks ever smaller and more localized.
> Mine is “the healthcare system doesn’t fall apart”.
Post-failure strategy.
> [Describing how the vaccine could have been rolled out more effectively.]
Yes, it could have been. But even so, by now almost everyone has been immunized either by the vaccine or prior infection and it's still killing huge numbers.
> N95 masks aren’t necessarily comfortable and people will avoid wearing them even when there’s a mandate.
I don't advocate for a mandate. Mandates lead to malicious compliance and people rebelling whenever they can to prove they still have autonomy.
> Half the time I saw people with masks covering just the mouth.
Like that. Fortunately, N95 are thicker and less likely to droop.
> And at the end of the day, your eyes and nose are likely other vectors for infection (eg studies that showed that people with glasses had a statistically significant lower rate of infection).
Practically insignificant.
> What I’ve heard from people with kids is that kids are an amazing infection vector for the whole family for any disease, COVID included.
Yes, household transmission is still an issue. We need to guarantee that people with Covid in the household can quarantine without repercussions.
> I fully expect that even with an over abundance of masks, nothing actually meaningfully changes. You may disagree with that conclusion
I do.
> N95 would be nice, but any and all masks, especially if worn correctly and consistently, offer meaningful statistically significant reduction in transmission rates
The difference between N95 and a surgical mask is far more than statistically significant. (See your own link.)
> Evolution tells us the virus will have pressures from immune systems and vaccines so it’ll probably evolve (few viruses these days can be fully and permanent eliminated by a vaccine).
A virus filtered out before it can enter the host never has a chance to evolve.
> However, I did think that transmission rates will be reduced at least a little bit (statistically appears true although with high enough variance that it can feel false) and more importantly death rate from COVID will plummet (also true).
No argument on the "little bit".
>> Blatantly false. It's still killing several multiples of the flu. And now the flu is back on top of that.
> Huh? Aren’t you basically agreeing with me then?
No.
> To me, the flu returning and COVID running more rampant is a clear sign that people stopped masking and socially distancing
Well, I agree with that. Masks also help prevent the flu better than the vaccines, and we don't even need to develop new masks every year. Just produce enough of the right types of masks that they're free/effectively free so they can do the bulk of the work, and all the other layers of defense can have the best foundation to start from.
Whew, that was a tiring Gish Gallop. Got any more?
... don't exist. There is still not enough production of the types of masks that are known and shown to be effective.
> even if you don’t wear it and are near someone with COVID you’re less likely to get it.
That was the hope, but it's not nearly as true as it needs to be.
> And when you do, you weather it significantly better.
And continue spreading it.
> So you need to keep up to date with shots.
And keep up with producing and distributing N95 masks to actually solve the problem.
> With a pandemic (now epidemic)
Wordplay with no bearing on reality.
> you have to deploy multiple measures (masks, vaccines, social distancing)
Sure. Defense in depth. Vaccines, especially vaccines that grossly fail to live up to expectations, aren't enough.
> pre-COVID is the same as post-Spanish flu
Blatantly false. It's still killing several multiples of the flu. And now the flu is back on top of that.
> COVID is here to stay permanent I think.
Probably true. People are stupid.